Test that getsolution is feasible when problem is Unbounded

[blegat]

LP

Solver	First test	Second test
Clp	[0, 0]	Pass
GLPK	Pass	Pass
CPLEX	Pass	[0,0]
Gurobi v6.51	Pass	[6.937e-310,0]
FICO XPress	No	License :(

Note: For CPLEX I had to set CPX_PARAM_REDUCE to 1. Note: For both Gurobi and Clp, the ticket number is 80, coincidence ?

Conic

Solver	First test	Second test
CSDP	[1.0,-4.22555e-7]	[4.63449e-9,-1.0]
Mosek	No solution	available error

[mlubin]

Which solvers currently pass this test?

[blegat]

I don't think that we can impose conic solver to support this so those tests should be in linprogsolvertestextra. @mlubin @joehuchette why no solver run linprogsolvertestextra in their tests ?

[mlubin]

why no solver run linprogsolvertestextra in their tests ?

No good reason

[mlubin]

It's troubling that neither CPLEX nor Gurobi pass both tests. I'd like to know if this behavior is reasonable to expect from an LP solver before imposing it via tests.

[blegat]

I have asked Gurobi whether the X attribute is supposed to be feasible when the problem is unbounded. We'll see what they say.

[chriscoey]

relevant to new status discussion https://github.com/JuliaOpt/MathProgBase.jl/issues/164 which should be resolved soon